endless delegreturn recovery
After a server reboot, the client endlessly tries to return a delegation to the server. The server replies with STALE_STATEID, which just triggers a new recovery cycle on the client.
This blocks unmount, and in case of a mirror mount, all new mounts on the server. Also it generates a load proportional to the number of mounts to the server, in our case the load is about 20, which renders the system barely usable.
Also, it prevents the regular open-reclaim from finishing in time (during the grace period), resulting in lost writes and locks.
No data to display